Microsoft Prepares the Launch of an Open-Source LLM for Math Tutoring

IBL News | New York

Microsoft announced it is exploring AI-powered math tutoring with Khan Academy using a new open-source small language model.

The model is a fine-tuned version of Phi-3, Microsoft’s family of small language models (SLMs). It has been trained on data from Accelerate Learning and UPchieve, including synthetic math tutoring chats.

In this initiative, Khan Academy provides explanatory educational content, including high-quality math problem questions and step-by-step answer sets, ongoing feedback, and benchmarking data to evaluate performance.

This tutoring tool can generate problems for specific grade levels and math topics, help students learn through guided explanations, and provide customized summaries of tutoring sessions for students and teachers with initial functionality supporting middle-school math.

“Phi-3 models not only outperform models of the same size and next size up across language, reasoning, coding, and math benchmarks, but they also do so in more cost-effective ways,” said Microsoft.

Preview of phi-3-vision-128k-instruct with NVIDIA

Graphic showing Phi-3 models compare to other models of similar size.